GenMentor LLM-powered & Goal-oriented Tutoring System

GenMentor is an open-source, LLM-powered multi-agent tutoring system for goal-oriented learning: it maps a learner's goal to the skills required, schedules a personalized learning path and generates tailored content. Accepted to WWW 2025 (Industry Track) as an oral presentation. Try it in your browser at genmentor.aurax.live.

Tianfu Wang1, Yi Zhan2, Jianxun Lian3, Zhengyu Hu1, Nicholas Jing Yuan2,†, Qi Zhang2, Xing Xie3, Hui Xiong1,†

1Hong Kong University of Science and Technology (Guangzhou) 2Microsoft 3Microsoft Research Asia †Corresponding authors

WWW 2025 (Industry Track) · Oral Presentation

Live Demo Paper Code Video ACM DL

GenMentor delivers goal-oriented learning by identifying each learner's knowledge gap and tailoring the learning path to meet their individual goal with precision.

0.67 Goal-to-skill recall, +59% over the best prompting baseline
4.71/5 Learning-path engagement rated by learners in the evaluation
3 Adaptive agents for personalized tutoring

01 · Overview

Abstract

Intelligent Tutoring Systems (ITSs) have revolutionized education by offering personalized learning experiences. However, as goal-oriented learning, which emphasizes efficiently achieving specific objectives, becomes increasingly important in professional contexts, existing ITSs often struggle to deliver this type of targeted learning experience.

In this paper, we propose GenMentor, an LLM-powered multi-agent framework designed to deliver goal-oriented, personalized learning within ITS. GenMentor begins by accurately mapping learners' goals to required skills using a fine-tuned LLM trained on a custom goal-to-skill dataset. After identifying the skill gap, it schedules an efficient learning path using an evolving optimization approach, driven by a comprehensive and dynamic profile of learners' multifaceted status. Additionally, GenMentor tailors learning content with an exploration-drafting-integration mechanism to align with individual learner needs.

Extensive automated and human evaluations demonstrate GenMentor's effectiveness in learning guidance and content quality. Furthermore, we have deployed it in practice and also implemented it as an application. A practical human study with professional learners further highlights its effectiveness in goal alignment and resource targeting, leading to enhanced personalization.

02 · Motivation

Goal-oriented Learning

Why professional learners need a tutor that starts from the goal, not from the syllabus.

Limitations of Traditional ITS

Existing ITSs rely on static curricula and reactive responses, lacking the ability to proactively guide learners toward specific professional objectives.

Professional Learning Needs

Modern professionals acquire targeted skills for specific goals; without clear guidance, the motivation and efficiency of learners may decline.

Goal-oriented Learning

Our approach emphasizes efficient goal achievement through personalized skill identification, adaptive learning paths, and targeted content delivery.

Evolution of ITS Paradigms

Diagram comparing traditional ITS, LLM-based ITS and goal-oriented ITS
Traditional ITS LLM-based ITS Goal-oriented ITS

03 · Architecture

GenMentor Framework

Three cooperating agents turn a goal into a skill gap, a schedule and tailored content.

GenMentor framework overview: skill gap identifier, adaptive learner modeler and personalized resource delivery

Skill Gap Identification

Dataset Construction

  • Custom goal-to-skill dataset
  • Built from job postings
  • Chain-of-thought reasoning tracks

LLM Fine-tuning

  • Goal-aligned skill mapping
  • Improved accuracy and relevance
  • Enhanced goal alignment

Gap Analysis

  • Skill competency assessment
  • Systematic gap identification
  • Skill and level gap as output

04 · Evaluation

Experimental Results

LLM-based automated evaluation of each component, plus an end-to-end study with professional learners.

LLM-based automated evaluation results for skill identification, learning path and content quality

Key performance highlights

Skill Identification

  • High accuracy in goal-to-skill mapping
  • Effective gap analysis

Learning Path Quality

  • Optimal progression structure
  • Strong learner engagement

Content Generation

  • High relevance to goals
  • Effective personalization

05 · Demo

Demo Video

Try GenMentor in your browser at genmentor.aurax.live, or run the full system locally by following the README quick-start.

06 · FAQ

Frequently Asked Questions

What is GenMentor?

GenMentor is an LLM-powered multi-agent framework for goal-oriented learning in intelligent tutoring systems (ITS). It maps a learner's goal to the skills required, identifies the skill gap, schedules a personalized learning path and generates tailored content. The work was accepted to the WWW 2025 Industry Track as an oral presentation, and the code is open source on GitHub.

How is GenMentor different from a chatbot tutor or a MOOC?

A MOOC follows a static syllabus and a chatbot tutor answers questions reactively. GenMentor is proactive: it starts from the learner's goal, plans the path to reach it and adapts that path as a dynamic learner profile evolves, so every resource is tied to a concrete objective.

Which agents make up the framework?

A Skill Gap Identifier, built on an LLM fine-tuned with a custom goal-to-skill dataset; an Adaptive Learner Modeler that maintains a multifaceted learner profile; a Learning Path Scheduler driven by an evolving optimization approach; a Tailored Content Generator that uses an exploration-drafting-integration mechanism; and an AI tutor chatbot for in-context help.

How was GenMentor evaluated?

Through LLM-based automated evaluation of skill identification, learning-path quality and content generation, plus an end-to-end user study with professional learners. In the automated evaluation, goal-to-skill recall reached 0.67, and learners rated learning-path engagement 4.71 out of 5.

Can I try or run GenMentor myself?

Yes. A live demo runs at genmentor.aurax.live with nothing to install. To run it locally, follow the repository's README quick-start and supply your own LLM API key (OpenAI and DeepSeek are supported out of the box). A lightweight Streamlit preview with limited functionality is also hosted at gen-mentor.streamlit.app.

07 · Citation

Citation

Published in Companion Proceedings of the ACM on Web Conference 2025 (WWW '25), Sydney.
DOI 10.1145/3701716.3715244 · arXiv:2501.15749 

@inproceedings{www-2025-genmentor,
  title     = {LLM-powered Multi-agent Framework for Goal-oriented Learning in Intelligent Tutoring System},
  author    = {Wang, Tianfu and Zhan, Yi and Lian, Jianxun and Hu, Zhengyu and Yuan, Nicholas Jing and Zhang, Qi and Xie, Xing and Xiong, Hui},
  booktitle = {Companion Proceedings of the ACM on Web Conference 2025 (WWW Companion '25)},
  year      = {2025},
  publisher = {ACM},
  doi       = {10.1145/3701716.3715244}
}